Abstract: [Purpose/Significance] Policy tools are the means and tools for government to implement govern targets and visions, which are important research domain for policy analysis. Considering that the current analysis of policy tools is still in the stage of manual classification, and there are a series of problems such as inconsistent standard of coding check, hard to reproduce, small scale, high cost and so on, this paper proposes to build an automatic classification model of policy tools. [Design/Methodology] This paper systematically sorted out the existing policy tool classification framework, and on the basis of Rothwell and Zegveld policy tool classification system,proposed an automatic policy tool classification model based on WordBERT and BiLSTM.Taking the the datasets of data governance and digital economy policy as an example,we independently constructed the data set and carried out three sets of experiments to verify the advantages and disadvantages of the model. [Findings/Conclusion] We find the automatic classification model of policy tools proposed in this paper work best, with a precision of 73.91%, which provide a highly efficient automatic classification method for the tedious work of policy tools classification. [Originality/Value] Aiming at the difficult problem of automatic classification of policy tools, this paper proposes to utilize unsupervised representation learning and supervised machine learning algorithms for automatic classification, so as to provide a strong tool for policiometrics analysis.